The Deliverator – Wannabee

So open minded, my thoughts fell out…

The Last Hard Drive I Will Ever Buy?

Posted by Deliverator on September 23rd, 2009

I needed to free up a fair sized hard drive to donate to Minimus and simultaneously had pretty much filled up my main content storage drive on my main desktop PC, so I went to Frys and picked up a Seagate 2 Terabyte drive and proceeded to spend a few days copying data back and forth between my main desktop’s many hard drives in order to free up a 500 GB drive. This is a process I have done MANY times over the years. I still have some data from the first hard disk my family owned, which held a whopping 80 MB (that’s megabytes folks). While doing all this, I realized I had a pretty good cross section of drives manufactured in the last 5 years in my personal possession and couldn’t resist doing some quick benchmarking. I’ve posted a gallery full of the results here and thought I would make a few quick comments on notable trends.

One of the results which is immediately apparent after viewing the results is that as drive capacities have gone up, sequential read speeds have gone up as well. This makes sense, as greater bit density on the platters means that more data passes under the read/write heads for a given unit of arc.

At the same time, random access times have gotten progressively worse.  Using a lower level drive utility like Spinrite on today’s ultra high capacity drives immediately reveals the reason for this trend – today’s drives are having an ever greater problem with seeking to the correct location over the platter in a reasonable amount of time. Spinrite shows a constant barrage of head seeking errors and reliance on error correcting code even with its extremely sequential access patterns. 12 milliseconds used to be a fairly typical random access time on a 7200 RPM hard drive. This has now gone up to more like 15 milliseconds on high density platter drives. My new 2 TB drive actually spins at a mere 5900 RPM, likely because head seeking errors were too high at 7200 RPM. This has lead to the extremely weird precedent of slowing down a drive to increase real world performance.

Real world performance is heavily dictated by random i/o patterns (particularly on fragmented hard drives).  This has created a niche market for lower capacity drives with high spindle speed and low random seek times, such as the Western Digital Raptor (now Velociraptor) line of drives. An older 74 GB model of which can be seen on the results page. This drive easily bests all the other conventional hard drives I’ve tested in terms of random access times (8 ms)  and I can attest to the fact that it basically never needs to re-seek. So, for years now performance enthusiasts have mixed and matched drives in their systems, using drives like those in the WD Raptor line for their main OS and program storage and using huge , poor performing drives for bulk storage of content such as movies, music and photos.

All this discussion is basically moot, as the one solid state disk I tested easily blows every hard disk I’ve ever used away in terms of performance. The OCZ Agility 64 GB SSD, which isn’t a particularly high end SSD, delivered sequential transfer rates 50-100% better than any conventional hard drive tested and random access times soo low I am not sure the benchmark tool even properly measured them (.1 ms). The effect this low random access time has on real world app performance is huge. We are talking Windows cold boot times measured in seconds here.

The vast majority of PC users I’ve encountered in my consulting tend to have under 30 GB of data. For these users, I see no reason for them to ever use a conventional hard disk ever again. Conventional hard disks, in my mind, should today be relegated to bulk storage and backup purposes only and by the time my giant 2 TB hdd fills up, I expect there to be equivalent size SSD equivalents available. It may very well be the last hard disk I ever buy and all I can say is ABOUT TIME.